Method, apparatus, device, and storage medium for splitting an image

Abstract

This application is applicable to the field of image processing technology, and provides a method, device, equipment and storage medium for segmenting images, including: obtaining a T1-weighted image of the abdomen of the object to be measured; using a trained image segmentation model to perform removal processing on the image corresponding to the liver blood vessels in the T1-weighted image to obtain a liver segmentation image. In the above solution, by processing the T1-weighted image with a trained image segmentation model, the image corresponding to the liver blood vessels in the T1-weighted image is removed, and without the interference of the liver blood vessels, the obtained liver segmentation image is more accurate.

Description

Technical Field

[0001] This application belongs to the technical field of image processing, and particularly relates to a method, apparatus, device, and storage medium for segmenting images. Background Art

[0002] Non-alcoholic fatty liver disease (NAFLD) is a clinicopathological syndrome characterized by excessive intrahepatic fat deposition excluding alcohol and other definite liver-damaging factors, and is an acquired metabolic stress-induced liver injury closely related to insulin resistance and genetic susceptibility. It mainly includes simple fatty liver (SFL), non-alcoholic steatohepatitis (NASH), and its related cirrhosis.

[0003] With the improvement of living standards and the change of lifestyle, the incidence of NAFLD is increasing, and it may cause liver cancer when severe. Therefore, it is crucial to perform accurate quantitative analysis of NAFLD at an early stage.

[0004] The accurate quantitative analysis of NAFLD is closely related to whether the fat quantification result (the fat content of the liver) is accurate, and whether the fat quantification result is accurate is in turn closely related to whether the liver image can be accurately segmented from the magnetic resonance (MR) image of the subject's abdomen.

[0005] In the prior art, a multi-layer perceptron (MLP) and a watershed algorithm are usually used to segment the MR image to obtain the liver image. However, the liver image obtained by using this method is not accurate. Summary of the Invention

[0006] In view of this, the embodiments of this application provide a method, apparatus, device, and storage medium for segmenting images to solve the problem that the liver image obtained by segmenting the MR image in the prior art is not accurate.

[0007] The first aspect of the embodiments of this application provides a method for segmenting images, and the method includes:

[0008] Obtain a T1-weighted image of the abdomen of the subject to be measured;

[0009] Through a trained image segmentation model, perform a removal process on the image corresponding to the liver blood vessels in the T1-weighted image to obtain a liver segmentation image.

[0010] In the above solution, by processing the T1-weighted image through a trained image segmentation model, the image corresponding to the liver blood vessels in the T1-weighted image is removed. Without the interference of the liver blood vessels, the obtained liver segmentation image is more accurate.

[0011] Optionally, the T1-weighted images include a T1-weighted in-phase image and a T1-weighted opposed-phase image. The method of obtaining a liver segmentation image by removing the image corresponding to the liver blood vessels in the T1-weighted images through the trained image segmentation model includes:

[0012] Extracting the liver edge features in the T1-weighted opposed-phase image and the liver blood vessel features in the T1-weighted in-phase image through the image segmentation model;

[0013] Determining a liver image according to the liver edge features;

[0014] Determining the image corresponding to the liver blood vessels from the liver image according to the liver blood vessel features;

[0015] Removing the image corresponding to the liver blood vessels from the T1-weighted images to obtain the liver segmentation image.

[0016] Optionally, after obtaining the liver segmentation image by removing the image corresponding to the liver blood vessels in the T1-weighted images through the trained image segmentation model, the method further includes:

[0017] Obtaining a proton density fat fraction quantitative image of the abdomen of the subject to be measured;

[0018] Registering the liver segmentation image and the proton density fat fraction quantitative image to obtain a fat quantification result.

[0019] Optionally, the training method of the trained image segmentation model is:

[0020] Training an initial model by using a preset sample training set and a loss function to obtain the trained image segmentation model. The sample training set includes a plurality of sample T1-weighted images and the liver mask images corresponding to each of the sample T1-weighted images. The liver mask image is an image with the liver region marked and the liver blood vessels removed.

[0021] Optionally, the training of the initial model by using a preset sample training set and a loss function to obtain the trained image segmentation model includes:

[0022] Inputting the sample T1-weighted images in the sample training set into the initial model for processing to obtain sample segmentation images of the sample T1-weighted images;

[0023] Calculating a loss value between the sample segmentation image of the sample T1-weighted image and the liver mask image corresponding to the sample T1-weighted image according to the loss function;

[0024] When it is detected that the loss value is greater than a preset threshold, adjust the model parameters of the initial model, and continue to train the initial model using the sample training set;

[0025] When it is detected that the loss value is less than or equal to the preset threshold, stop training the initial model, and determine the trained initial model as the image segmentation model.

[0026] Optionally, training the initial model using a preset sample training set and a loss function to obtain the trained image segmentation model includes:

[0027] Input the sample T1-weighted image in the sample training set into the initial model for processing to obtain a sample segmentation image of the sample T1-weighted image;

[0028] Calculate the loss value between the sample segmentation image of the sample T1-weighted image and the liver mask image corresponding to the sample T1-weighted image according to the loss function;

[0029] Output an intermediate model whenever it is detected that the loss value is less than or equal to the preset threshold until the number of training times reaches the preset number of training times to obtain multiple intermediate models;

[0030] Determine the image segmentation model from the multiple intermediate models.

[0031] Optionally, determining the image segmentation model from the multiple intermediate models includes:

[0032] Obtain a sample test set, where the sample test set includes multiple test T1-weighted images and liver mask test images corresponding to each of the test T1-weighted images;

[0033] Test the multiple intermediate models using the sample test set;

[0034] Determine the image segmentation model from the multiple intermediate models according to each test result.

[0035] A second aspect of the embodiments of the present application provides a device for segmenting an image, including:

[0036] An acquisition unit, configured to acquire a T1-weighted image of the abdomen of a measured object;

[0037] A processing unit, configured to remove the image corresponding to the liver blood vessels in the T1-weighted image through a trained image segmentation model to obtain a liver segmentation image.

[0038] The third aspect of the embodiments of the present application provides a device for segmenting images, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the method for segmenting images as described in the first aspect above are implemented.

[0039] The fourth aspect of the embodiments of the present application provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps of the method for segmenting images as described in the first aspect above are implemented.

[0040] The fifth aspect of the embodiments of the present application provides a computer program product, and when the computer program product runs on a device for segmenting images, the device for segmenting images is caused to execute the steps of the method for segmenting images as described in the first aspect above. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only some embodiments of the present application, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0042] Figure 1 is a schematic flowchart of a method for segmenting images provided by an exemplary embodiment of the present application;

[0043] Figure 2 is a specific flowchart of step S102 of a method for segmenting images shown in another exemplary embodiment of the present application;

[0044] Figure 3 is a liver-segmented image provided by the present application;

[0045] Figure 4 is a schematic flowchart of a method for segmenting images shown in another exemplary embodiment of the present application;

[0046] Figure 5A and Figure 5B is a comparison diagram provided by the present application;

[0047] Figure 6 is a schematic diagram of a device for segmenting images provided by an embodiment of the present application;

[0048] Figure 7 is a schematic diagram of a device for segmenting images provided by another embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] In order to make the objectives, technical solutions and advantages of this application more clear and understandable, the following further details this application in combination with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit this application.

[0050] Non-alcoholic fatty liver disease (NAFLD) refers to a clinical and pathological syndrome characterized by excessive fat deposition in liver cells excluding alcohol and other clear liver-damaging factors, and is an acquired metabolic stress-induced liver injury closely related to insulin resistance and genetic susceptibility. It mainly includes simple fatty liver (SFL), non-alcoholic steatohepatitis (NASH) and its related cirrhosis.

[0051] With the improvement of living standards and the change of lifestyle, the incidence of NAFLD is getting higher and higher. According to the survey, the prevalence of NAFLD in the whole country is about 29.2%, and it can cause liver cancer when severe. Therefore, it is crucial to perform precise quantitative analysis of NAFLD at an early stage.

[0052] The precise quantitative analysis of NAFLD is closely related to whether the fat quantification result (the fat content of the liver) is accurate. In actual clinical applications, it is necessary to take a biopsy sample of the liver of the subject or obtain a liver image through imaging examination to further evaluate and diagnose. Liver biopsy is the gold standard for the diagnosis of NAFLD, but it is highly subjective and has problems such as invasiveness, high risk, sampling error and non-repeatability; ultrasound examination is non-invasive, economical and simple, but it is a non-quantitative examination, with poor clarity and resolution, and is not sensitive to mild patients; CT imaging (a technology that combines X-ray scan projection data with reconstruction mathematics and computer technology to obtain medical images based on slice information) has disadvantages such as ionizing radiation, insensitivity to mild fatty liver and low quantitative accuracy. Therefore, they are not suitable for liver fat quantitative analysis.

[0053] Magnetic resonance imaging technology has become a powerful means for liver fat quantitative analysis due to its advantages such as multi-parameter imaging, high resolution and no ionizing radiation. Its unique proton density fat fraction (PDFF) quantitative image can accurately reflect the fat proportion in liver cells and is highly consistent with the pathological results.

[0054] The PDFF quantitative image is determined by using magnetic resonance chemical shift encoded fat quantification technology. Specifically, this technology utilizes the resonance frequency difference between water protons and fat protons in tissues, adopts a multi-echo gradient echo imaging sequence to collect magnetic resonance signals with different water-fat phase differences, calculates the pure water signal and pure fat signal based on the water-fat separation algorithm, and thereby quantitatively determines the proportion of fat volume in liver tissue to obtain the PDFF quantitative image.

[0055] Collect magnetic resonance (MR) images of the abdomen of the subject to be examined, segment the liver image from the MR images, and calculate the fat quantification result (the fat content of the liver) of the subject based on the MR images and PDFF quantitative images. Therefore, the accuracy of the fat quantification result is closely related to whether the liver image can be accurately segmented from the MR images of the subject's abdomen.

[0056] The traditional region growing segmentation algorithm merges similar pixel points within the range of the adjacent region of the seed point according to the set seed point and region growing criterion, and segments out the connected regions with the same characteristics to obtain the liver image. However, this method lacks robustness for liver images with uneven gray levels or containing noise, which will lead to holes or under-segmentation in the segmentation results and is not suitable for segmenting liver images with complex backgrounds.

[0057] Using a multi-layer perceptron (MLP) and a watershed algorithm to jointly segment the MR images can also obtain the liver image. However, this method is prone to phenomena such as blurred edges, under-segmentation (the liver region is not completely segmented), and mis-segmentation (regions that do not belong to the liver are segmented as the liver).

[0058] Moreover, whether it is the traditional segmentation method or the improved segmentation method, the segmented liver images contain a large number of images of liver blood vessels, resulting in inaccurate segmented liver images. And due to the interference of liver blood vessels in the liver image, the finally calculated fat content value will be less than the true fat content value. Therefore, excluding the images of liver blood vessels in the liver image is extremely crucial for the accuracy of the fat quantification result.

[0059] In view of this, the embodiment of the present application provides a method for segmenting images. First, obtain the T1-weighted image of the abdomen of the object to be measured; through the trained image segmentation model, perform a removal process on the image corresponding to the liver blood vessels in the T1-weighted image to obtain the liver segmentation image. In this method, by processing the T1-weighted image through the trained image segmentation model, the image corresponding to the liver blood vessels in the T1-weighted image is removed, making the obtained liver segmentation image more accurate. And without the interference of liver blood vessels, the finally calculated fat content value based on this liver image can be closer to the true fat content value, thereby improving the accuracy of the fat quantification result.

[0060] Please refer to Figure 1 , Figure 1FIG. 0 is a schematic flowchart of a method for segmenting an image provided by an exemplary embodiment of the present application. The execution subject of the method for segmenting an image provided by the present application is a device for segmenting an image, where the device includes, but is not limited to, an in-vehicle computer, a tablet computer, a computer, a personal digital assistant (PDA), etc., and may also include various types of servers. For example, the server can be an independent server or a cloud service that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery network (CDN), and big data and artificial intelligence platforms.

[0061] As Figure 1 shown, the method for segmenting an image may include: S101 - S102, specifically as follows:

[0062] S101: Obtain a T1-weighted image of the abdomen of the object to be measured.

[0063] The T1-weighted image refers to the MR image of the abdomen of the object to be measured collected, which mainly reflects the difference in T1 values between tissues. Among them, T1 represents the longitudinal relaxation time (the time for the longitudinal magnetic vector to recover).

[0064] Exemplarily, the MR image of the abdomen of the object to be measured is collected by a magnetic resonance scanner. The MR image has multiple imaging parameters, that is, the MR image has a T1 value reflecting the T1 relaxation time, a T2 value reflecting the T2 relaxation time, and a relaxation time value reflecting the proton density, etc. If the MR image mainly reflects the difference in T1 values between tissues, it is a T1-weighted image (T1WI); if it mainly reflects the difference in T2 values between tissues, it is a T2-weighted image (T2WI); if it mainly reflects the difference in proton density relaxation time between tissues, it is a proton density weighted image (PDWI).

[0065] In this embodiment, the collected MR image mainly reflects the difference in T1 values between tissues, that is, a T1-weighted image.

[0066] Specifically, in practical applications, the device for segmenting an image is pre-connected to the magnetic resonance scanner for communication. After the magnetic resonance scanner collects the T1-weighted image, it sends the collected T1-weighted image to the device.

[0067] For example, the magnetic resonance scanning parameters of a magnetic resonance scanner can be set, and then the magnetic resonance scanner is controlled to scan the abdomen of the object to be measured, so that the magnetic resonance scanner acquires the T1-weighted image of the abdomen of the object to be measured. The magnetic resonance scanner sends the acquired T1-weighted image to the device, and the device receives the T1-weighted image sent by the magnetic resonance scanner.

[0068] In a possible implementation manner of the embodiment of the present application, the magnetic resonance scanning parameters can be set to a three-dimensional magnetic resonance FLASH sequence for T1-weighted image acquisition. For example, the scanning field of view is set to 280×400 mm2, the in-plane resolution is set to 2.08×2.08 mm2, the slice thickness is set to 6 mm, the flip angle is set to 12°, etc. This is only for illustrative purposes and is not limited thereto.

[0069] Exemplarily, the T1-weighted image may include a T1-weighted in-phase image and a T1-weighted opposed-phase image.

[0070] The human magnetic resonance signal mainly comes from water and fat. The precession frequency of hydrogen protons in water is faster than that of hydrogen protons in fat. For the hydrogen protons in water and fat in the same pixel, after RF excitation, the transverse magnetization vectors of water and fat are in the same phase. When the RF stops, since the precession frequency of hydrogen protons in water is faster than that of hydrogen protons in fat, after a period of time, when the phase difference between the hydrogen protons in water and the hydrogen protons in fat is 180 degrees, their macroscopic transverse magnetization vectors cancel each other out, and the signal detected by the magnetic resonance scanner is the difference between the water and fat signals. Such an image is called an opposed-phase image.

[0071] After another period of time, when the phase difference between the hydrogen protons in water and the hydrogen protons in fat is 360 degrees, their phases coincide again. At this time, the signal detected by the magnetic resonance scanner is the sum of the water and fat signals. Such an image is called an in-phase image.

[0072] During the relaxation process, the phases of the above two repeat cyclically, and in-phase and opposed-phase will appear in turn, so that the T1-weighted in-phase image and the T1-weighted opposed-phase image of the abdomen of the object to be measured can be acquired.

[0073] S102: Through the trained image segmentation model, the image corresponding to the liver blood vessels in the T1-weighted image is removed to obtain a liver segmentation image.

[0074] The trained image segmentation model is based on the Pyramid Attention Network (PAN) and is obtained by training the PAN using a sample training set. The sample training set includes multiple sample T1-weighted images and the corresponding liver mask images for each sample T1-weighted image. Among them, the sample T1-weighted images include sample T1-weighted in-phase images and sample T1-weighted opposed-phase images. The liver mask image is an image marked with the liver area and with the liver blood vessels removed.

[0075] In this embodiment, the image segmentation model can be pre-trained by this device or can be pre-trained by other devices and then the file corresponding to the image segmentation model is transplanted into this device. That is to say, the execution entity for training the image segmentation model and the execution entity for performing image segmentation using the image segmentation model can be the same or different. For example, when using other devices to train the image segmentation model, after other devices finish training the image segmentation model, the model parameters of the image segmentation model are fixed to obtain the file corresponding to the trained image segmentation model, and then this file is transplanted into this device.

[0076] By using the trained image segmentation model to perform segmentation processing on the T1-weighted image, the image corresponding to the liver in the abdomen of the measured object can be segmented out, and the image corresponding to the liver blood vessels in the T1-weighted image is removed to obtain the final liver segmentation image.

[0077] In this embodiment, by using the trained image segmentation model to process the T1-weighted image, the image corresponding to the liver blood vessels in the T1-weighted image is removed, making the obtained liver segmentation image more accurate. And without the interference of the liver blood vessels, the fat content value finally calculated based on this liver image can be closer to the true fat content value, thereby improving the accuracy of the fat quantification result.

[0078] Please refer to Figure 2 , Figure 2 which is the specific flowchart of step S102 of a method for segmenting an image shown in another exemplary embodiment of this application; optionally, in some possible implementation manners of this application, the above S102 may include S1021 to S1024, specifically as follows:

[0079] S1021: Extract the liver edge features in the T1-weighted opposed-phase image and the liver blood vessel features in the T1-weighted in-phase image through the image segmentation model.

[0080] Exemplarily, the trained image segmentation model may include a Feature Pyramid Attention (FPA) module and a Global Attention Upsampling (GAU) module. The FPA module and the GAU module can be used to extract liver edge features in the T1-weighted opposed-phase image and extract liver vascular features in the T1-weighted in-phase image.

[0081] Exemplarily, the FPA module contains multiple convolutional layers, such as 7*7 convolutional layer, 5*5 convolutional layer, 3*3 convolutional layer. Through the multiple convolutional layers in the FPA module, downsampling operation is performed on the T1-weighted opposed-phase image to extract the liver edge features corresponding to the liver in the T1-weighted opposed-phase image. This liver edge feature can be used to indicate the boundary of the liver.

[0082] The GAU module also contains convolutional layers, such as 3*3 convolutional layer. Through the convolutional layer in the GAU module, convolutional operation is performed on the T1-weighted in-phase image to extract the liver vascular features corresponding to the liver in the T1-weighted in-phase image.

[0083] Optionally, in some possible implementation manners of the present application, before performing step S1021, preprocessing may also be performed on the T1-weighted opposed-phase image and the T1-weighted in-phase image. Then, the liver edge features in the preprocessed T1-weighted opposed-phase image and the liver vascular features in the preprocessed T1-weighted in-phase image are extracted through the image segmentation model.

[0084] Exemplarily, the preprocessing may include any one or a combination of normalization processing, cropping processing, and denoising processing. For example, the T1-weighted opposed-phase image and the T1-weighted in-phase image are cropped so that the sizes of the T1-weighted opposed-phase image and the T1-weighted in-phase image meet the preset sizes.

[0085] For another example, denoising processing is performed on the T1-weighted in-phase image to obtain a denoised image, and normalization processing is performed on the denoised image so that the pixel value corresponding to each pixel point in the denoised image falls within [0, 255] to obtain the preprocessed T1-weighted in-phase image. Or, normalization processing is directly performed on the T1-weighted in-phase image to obtain the preprocessed T1-weighted in-phase image. Similarly, preprocessing is also performed on the T1-weighted opposed-phase image to obtain the preprocessed T1-weighted opposed-phase image. This is only for exemplary illustration and is not limited thereto.

[0086] Performing preprocessing on the T1-weighted opposed-phase image and the T1-weighted in-phase image eliminates interference, makes the images more standardized, is beneficial to improving the speed of the image segmentation model to extract their respective corresponding features subsequently, and helps to improve the accuracy of the subsequent segmented images.

[0087] S1022: Determine the liver image according to the liver edge features.

[0088] Exemplarily, the image segmentation model generates an edge feature map based on the extracted liver edge features, and uses the GAU module to perform an upsampling operation on the edge feature map to restore the spatial information and edge information corresponding to the liver, obtaining a liver image that only contains the liver region.

[0089] S1023: Determine the image corresponding to the liver blood vessels from the liver image according to the liver blood vessel features.

[0090] Exemplarily, the image segmentation model performs a convolution operation on the extracted liver blood vessel features, thereby determining the image corresponding to the liver blood vessels in the liver image.

[0091] Optionally, the liver image is downsampled through multiple convolutional layers in the FPA module to extract the liver blood vessel edge features of the liver image. The liver blood vessel edge features can be used to indicate the boundary of the liver blood vessels.

[0092] According to the liver blood vessel features and the liver blood vessel edge features, determine the image corresponding to the liver blood vessels in the liver image. For example, combine the liver blood vessel features and the liver blood vessel edge features to obtain the image corresponding to the liver blood vessels. Specifically, the image segmentation model includes a pooling layer, and the pooling layer is used to connect the liver blood vessel features and the liver blood vessel edge features to obtain the image corresponding to the liver blood vessels in the liver image.

[0093] S1024: Remove the image corresponding to the liver blood vessels from the T1-weighted image to obtain a liver segmentation image.

[0094] Exemplarily, the T1-weighted image includes a T1-weighted in-phase image and a T1-weighted opposed-phase image. The image corresponding to the liver blood vessels can be removed from the T1-weighted in-phase image to obtain a liver segmentation image; or the image corresponding to the liver blood vessels can be removed from the T1-weighted opposed-phase image to obtain a liver segmentation image.

[0095] Specifically, removing the image corresponding to the liver blood vessels can be not displaying the image corresponding to the liver blood vessels. For example, mark each pixel point of the image corresponding to the liver blood vessels as 0, so that each pixel point of the image corresponding to the liver blood vessels is displayed in black, so that the image corresponding to the liver blood vessels is all displayed in black. Except for the image corresponding to the liver blood vessels in the liver image, all are displayed in white, thereby obtaining a liver segmentation image.

[0096] To more intuitively display the liver segmentation image, please refer to Figure 3 , Figure 3 is a liver segmentation image provided by the present application. As Figure 3 shown, Figure 3The left column shows three different T1-weighted in-phase images, the middle column shows three true liver images annotated by medical software, and the right column shows three liver segmentation images obtained by processing with an image segmentation model. For any liver segmentation image, the white area is the image corresponding to the segmented liver, and the black area in the white area is the image corresponding to the removed liver blood vessels.

[0097] From Figure 3 it can be clearly seen that the liver segmentation images obtained by processing with the image segmentation model in this application are extremely close to the standard true liver images. That is to say, the liver segmentation images obtained by processing with the image segmentation model in this application highly coincide with the actual liver and the area of the liver blood vessels of the measured object. That is, the accurate segmentation of the liver tissue with the liver blood vessels removed is achieved by the method in this application.

[0098] In this embodiment, features are extracted through the image segmentation model, the liver image is determined, and the image corresponding to the liver blood vessels is determined. Then, the image corresponding to the liver blood vessels is removed to obtain the liver segmentation image, realizing the accurate removal of the liver blood vessels in the liver tissue, thereby realizing the accurate segmentation of the liver and making the obtained liver segmentation image more accurate. Moreover, without the interference of the liver blood vessels, the fat content value finally calculated based on this liver image can be closer to the true fat content value, thus improving the accuracy of the fat quantification result.

[0099] Please refer to Figure 4 , Figure 4 which is a schematic flowchart of a method for segmenting an image shown in another exemplary embodiment of this application. As Figure 4 shown, a method for segmenting an image may include: S201 to S204, where S201 and S202 are exactly the same as S101 and S102 in the Figure 1 corresponding embodiment. For specific reference, see the description of S101 and S102 in the Figure 1 corresponding embodiment. S203 to S204 are specifically as follows:

[0100] S203: Obtain a proton density fat fraction quantitative image of the abdomen of the measured object.

[0101] Exemplarily, a proton density fat fraction quantitative image, that is, a PDFF quantitative image, can be obtained through magnetic resonance chemical shift encoding imaging technology.

[0102] There is a chemical shift between the hydrogen protons in water and the hydrogen protons in fat within human tissues. Using magnetic resonance chemical shift encoding imaging technology, magnetic resonance images with multiple different echo times are acquired using a magnetic resonance gradient echo imaging sequence. At different echo times, there are different phase differences between the water hydrogen protons and the fat hydrogen protons in the magnetic resonance images. Then, by solving through a preset encoding imaging model, the water hydrogen protons and the fat hydrogen protons can be separated to obtain a pure water image and a pure fat image. According to the proton density fat fraction = fat image / (fat image + water image), the proton density fat fraction can be obtained.

[0103] In practical applications, the device for segmenting images is pre-connected to a magnetic resonance scanner in communication. The magnetic resonance scanner acquires multi-echo images of the abdomen of the object to be measured and sends the acquired multi-echo images to the device. The device calculates the multi-echo images to obtain a PDFF quantitative image of the abdomen of the object to be measured.

[0104] It can also be that the magnetic resonance scanner acquires multi-echo images of the abdomen of the object to be measured, calculates the multi-echo images to obtain a PDFF quantitative image. The magnetic resonance scanner sends the PDFF quantitative image to the device, and the device receives the PDFF quantitative image sent by the magnetic resonance scanner.

[0105] For example, the magnetic resonance scanning parameters of the magnetic resonance scanner can be set, and then the magnetic resonance scanner is controlled to scan the abdomen of the object to be measured, so that the magnetic resonance scanner acquires multi-echo images of the abdomen of the object to be measured.

[0106] In a possible implementation manner of the embodiment of the present application, when using a magnetic resonance scanner to acquire a T1-weighted image of the abdomen of the object to be measured, multi-echo images of the abdomen of the object to be measured can be acquired simultaneously. Therefore, when acquiring the T1-weighted image and the multi-echo images, the parameters involved in both can be set to be the same. For example, the magnetic resonance scanning parameters are set to a three-dimensional magnetic resonance FLASH sequence, the scanning field of view is set to 280×400 mm2, the in-plane resolution is set to 2.08×2.08 mm2, and the slice thickness is set to 6 mm.

[0107] Since the multi-echo images use six-echo data acquisition, the repetition time can be set to 10.5 ms, and the echo times are respectively set to 1.67, 3.15, 4.63, 6.11, 7.59, 9.07 ms. This is only for illustrative purposes and is not limited thereto.

[0108] The method for calculating the PDFF quantitative image using multi-echo images can be to convert the multi-echo images into multi-echo complex-valued image data. Import the multi-echo complex-valued image data into the Iterative Decomposition of water and fat with Echo Asymmetry and Least-squares estimation (IDEAL) algorithm to solve the initial ratio of water and fat, which can be used to reconstruct the initial value of the water-fat ratio in terms of magnitude. Use the initial water-fat ratio as the initial value of the fitting regression model to obtain the estimated water-fat ratio after magnitude reconstruction and image. Divide the multi-echo complex-valued image data by the attenuation coefficient to obtain the complex-valued image data after attenuation correction. Import the corrected complex-valued image data into the IDEAL algorithm to obtain the estimated water-fat ratio after complex-valued reconstruction. Weight the estimated water-fat ratio, image, and the complex-valued image data after attenuation correction according to a preset method to obtain the final mixed-weighted PDFF image. The specific implementation process can refer to the prior art and will not be elaborated here.

[0109] S204: Register the liver segmentation image and the proton density fat fraction quantitative image to obtain the fat quantification result.

[0110] Use a registration method to register the liver segmentation image and the PDFF quantitative image to obtain the fat quantification result of the liver of the subject to be measured. Exemplarily, extract features from the liver segmentation image and the PDFF quantitative image to obtain the corresponding feature points of the liver segmentation image and the PDFF quantitative image respectively. Perform similarity measurement to find the matching feature point pairs among the corresponding feature points of the liver segmentation image and the PDFF quantitative image. Then, determine the image space coordinate transformation parameters through the matching feature point pairs. Finally, perform image registration by the coordinate transformation parameters to obtain the fat quantification result.

[0111] In this embodiment, use a registration method to register the liver segmentation image and the PDFF quantitative image to obtain the fat quantification result of the liver of the subject to be measured. Since the liver segmentation image obtained by using the segmentation method provided in this application excludes the image corresponding to the liver blood vessels and there is no interference from the liver blood vessels, the finally obtained fat quantification result is closer to the real result, that is, the liver fat content value is closer to the real fat content value, improving the accuracy of the fat quantification result and facilitating the accurate diagnosis of the subsequent subject to be measured.

[0112] Optionally, in a possible implementation manner, in order to more intuitively see the fat distribution of the liver of the subject to be measured, the fat quantification result includes an image of the liver fat content distribution. Please refer to Fig.Figure 5A and Figure 5B , Figure 5A and Figure 5B are the comparison diagrams provided by this application.

[0113] As Figure 5A and Figure 5B shown, Figure 5A on the left in is the result shown by the registered liver segmentation image stacked on the apparent transverse relaxation rate Figure 5A image, on the right in is the result shown by the unregistered liver segmentation image stacked on the apparent transverse relaxation rate

[0114] Figure 5B On the left in Figure 5B is the display diagram of the fat quantification result after removing the liver blood vessels, the mean is: 9.2441, and the median is: 8.6000.

[0115] On the right in Figure 5B is the display diagram of the fat quantification result without removing the liver blood vessels, the mean is: 8.8110, and the median is: 8.2000. Figure 5B The fat quantification result is obtained based on the registration of the liver segmentation image and the PDFF quantification image. The difference is that

[0116] Optionally, in some possible implementation manners of this application, a method for training an image segmentation model is further provided. Specifically, the training manner of the image segmentation model is: training an initial model by using a preset sample training set and a loss function to obtain a trained image segmentation model.

[0117] Exemplarily, the sample training set includes a plurality of sample T1-weighted images, and liver mask images corresponding to each of the sample T1-weighted images. The sample T1-weighted images may include sample T1-weighted in-phase images and sample T1-weighted opposed-phase images. The liver mask image is an image with the liver region marked and the liver blood vessels removed.

[0118] For example, T1-weighted images of the abdomens of multiple volunteers are collected by a magnetic resonance scanner as the sample T1-weighted images. To facilitate subsequent testing of the initial model during training, PDFF quantitative images corresponding to each T1-weighted image may also be collected, that is, PDFF quantitative images of the abdomens of multiple volunteers are collected. The collection method may refer to the description in S101 and will not be elaborated here.

[0119] The liver region is marked on the sample T1-weighted in-phase image and the liver blood vessels are removed, and the resulting image is the liver mask image. Specifically, the pixel points in the liver region of the sample T1-weighted in-phase image are set to 1, and the pixel points of the background, liver blood vessels, and other tissues are set to 0. After the setting is completed, the resulting image is the liver mask image. It should be noted that the sample T1-weighted in-phase image can be marked by means of manual marking, machine marking, or a combination thereof, etc., and there is no limitation in this regard.

[0120] Optionally, in order to increase the number of samples, the sample T1-weighted in-phase image, the sample T1-weighted opposed-phase image, and the liver mask image can be subjected to normalization processing, cropping processing, specified-angle rotation processing, flipping processing, etc., so as to obtain more samples.

[0121] Exemplarily, in practical applications, no less than 1000 groups of sample data can be selected as the sample training set. To improve the training effect, a part (such as 30% of the sample data) can be selected from the sample training set as the validation set for verifying the training effect of the image segmentation model during the process of training the image segmentation model.

[0122] The preset loss function may include a cross-entropy loss function, a logarithmic loss function, a mean square error formula, etc. The initial model may include a PAN network, a U-net network, etc. In this embodiment, the PAN network is taken as an example for illustration.

[0123] Exemplarily, the PAN network is trained using the preset sample training set and the cross-entropy loss function to obtain a trained image segmentation model.

[0124] In this embodiment, an initial model is trained using a preset sample training set and a loss function to obtain a trained image segmentation model. The liver mask image in the sample training set is an image with the liver region marked and the liver blood vessels removed. That is to say, this liver mask image is the target for the initial model to learn, and during the training process, the initial model will learn the features corresponding to the liver mask image. For example, for the liver mask image with the liver blood vessels removed, the initial model during training can learn that there are no pixel features at the position of the liver blood vessels, so that the finally trained image segmentation model can output a feature map with the blood vessels removed. This facilitates obtaining an accurate liver segmentation image when using this image segmentation model subsequently.

[0125] Optionally, in some possible implementation manners of the present application, training the initial model using a preset sample training set and a loss function to obtain a trained image segmentation model may specifically include: S301 - S304, as follows:

[0126] S301: Input the sample T1 - weighted image in the sample training set into the initial model for processing to obtain a sample segmentation image of the sample T1 - weighted image.

[0127] The untrained image segmentation model may include a PAN network, a U - net network, etc. In this embodiment, the PAN network is taken as an example for illustration.

[0128] Exemplarily, input the sample T1 - weighted image into the PAN network for processing to obtain a sample segmentation image corresponding to the sample T1 - weighted image. The structure of the PAN network is the same as that of the trained image segmentation model. The processing process of the PAN network for the sample T1 - weighted image can refer to the description in S1021 - S1024 and will not be elaborated here.

[0129] S302: Calculate the loss value between the sample segmentation image of the sample T1 - weighted image and the liver mask image corresponding to the sample T1 - weighted image according to the loss function.

[0130] Exemplarily, the preset loss function may include a cross - entropy loss function, a logarithmic loss function, a mean - square error formula, etc.

[0131] For example, calculate the loss value between the sample segmentation image of the sample T1 - weighted image and the liver mask image corresponding to this sample T1 - weighted image through the cross - entropy loss function.

[0132] When obtaining the loss value between the sample segmentation image of the sample T1 - weighted image and the liver mask image corresponding to the sample T1 - weighted image, determine the magnitude relationship between the loss value and a preset threshold. When the loss value is greater than the preset threshold, execute S303; when the loss value is less than or equal to the preset threshold, execute S304.

[0133] S303: When it is detected that the loss value is greater than a preset threshold, adjust the model parameters of the initial model, and continue to train the initial model using the sample training set.

[0134] Exemplarily, when the device performing the training process (for example, a device for segmenting images, or other devices) confirms that the current loss value is greater than the preset threshold, it is determined that the initial model currently being trained does not meet the requirements. At this time, adjust the model parameters (such as weight values) of the initial model being trained, and continue to train the initial model using the sample training set. That is, return to execute S301 - S302 until it is detected that the loss value is less than or equal to the preset threshold, and then execute S304.

[0135] S304: When it is detected that the loss value is less than or equal to the preset threshold, stop training the initial model, and determine the trained initial model as the image segmentation model.

[0136] Exemplarily, when the device performing the training process (for example, a device for segmenting images, or other devices) confirms that the current loss value is less than or equal to the preset threshold, it is determined that the initial model currently being trained meets the requirements. At this time, fix the model parameters in the initial model, and determine the initial model with the fixed model parameters as the trained image segmentation model.

[0137] In this embodiment, the image segmentation model is obtained after a large number of sample trainings, learns how to extract rich and effective feature information during the training process, and has a relatively small corresponding loss value. Therefore, using this image segmentation model to perform a removal process on the image corresponding to the liver blood vessels in the T1 - weighted image can obtain an accurate liver segmentation image.

[0138] Optionally, in some possible implementation manners of the present application, training the initial model using a preset sample training set and a loss function to obtain a trained image segmentation model may further include: S401 - S404, specifically as follows:

[0139] S401: Input the sample T1 - weighted image in the sample training set into the initial model for processing to obtain a sample segmentation image of the sample T1 - weighted image.

[0140] The untrained image segmentation model may include a PAN network, a U - net network, etc. In this embodiment, the PAN network is taken as an example for illustration.

[0141] Exemplarily, input the sample T1 - weighted image into the PAN network for processing to obtain a sample segmentation image corresponding to the sample T1 - weighted image. The structure of the PAN network is the same as that of the trained image segmentation model. The processing process of the PAN network for the sample T1 - weighted image can refer to the description in S1021 - S1024, and will not be elaborated here.

[0142] S402: Calculate the loss value between the sample segmentation image of the sample T1-weighted image and the liver mask image corresponding to the sample T1-weighted image according to the loss function.

[0143] Exemplarily, the preset loss function may include cross-entropy loss function, logarithmic loss function, mean square error formula, etc.

[0144] For example, calculate the loss value between the sample segmentation image of the sample T1-weighted image and the liver mask image corresponding to the sample T1-weighted image through the cross-entropy loss function.

[0145] S403: Output the intermediate model whenever the detected loss value is less than or equal to the preset threshold until the number of training times reaches the preset number of training times, and obtain multiple intermediate models.

[0146] The intermediate model is the initial model in the current training. The number of training times is the number of iterations, that is, the number of times to train the initial model. The preset number of training times is the preset number of iterations, and the preset number of training times can be set according to the actual situation and is not limited thereto.

[0147] Exemplarily, before the number of training times reaches the preset number of training times, the device executing the training process (for example, the device for segmenting images, or other devices) outputs an intermediate model whenever the detected loss value is less than or equal to the preset threshold during the training process. Until the number of training times reaches the preset number of training times, at which time multiple intermediate models are output.

[0148] For example, before the number of training times reaches the preset number of training times, the device for segmenting images continuously trains the initial model through the sample training set. During the training process, when the detected loss value is less than or equal to the preset threshold, an intermediate model is output, and then the intermediate model is continuously trained through the sample training set. When the detected loss value of the continuously trained intermediate model is less than or equal to the preset threshold, this intermediate model is output. And so on until the number of training times reaches the preset number of training times, at which time multiple intermediate models are obtained.

[0149] S404: Determine the image segmentation model from multiple intermediate models.

[0150] Exemplarily, the image segmentation model can be determined according to the loss values corresponding to each intermediate model. For example, obtain the loss values of each intermediate model, and determine the intermediate model with the smallest loss value as the trained image segmentation model.

[0151] Optionally, it can also be to test each output intermediate model to verify the quality of each intermediate model. Determine the image segmentation model from multiple intermediate models according to each test result.

[0152] In this embodiment, the initial model is trained with a sample training set to obtain multiple intermediate models, and the best intermediate model is selected from the multiple intermediate models as the trained image segmentation model, which ensures the quality of the image segmentation model. Subsequently, when using this image segmentation model to perform removal processing on the image corresponding to the liver blood vessels in the T1-weighted image, an accurate liver segmentation image can be obtained.

[0153] Optionally, in some possible implementation manners of the present application, the above S404 may include S4041 to S4043, which are specifically as follows:

[0154] S4041: Obtain a sample test set.

[0155] The sample test set includes multiple test T1-weighted images and the corresponding liver mask test images for each test T1-weighted image. The liver mask test image is an image with the liver region marked and the liver blood vessels removed.

[0156] Exemplarily, several sample T1-weighted images can be selected from the sample training set as the test T1-weighted images, and the corresponding liver mask images of the selected several sample T1-weighted images are used as the liver mask test images corresponding to each test T1-weighted image.

[0157] S4042: Test the multiple intermediate models using the sample test set.

[0158] Input the test T1-weighted images in the sample test set into each intermediate model for processing to obtain the test segmentation images of the test T1-weighted images. Obtain the PDFF quantitative image corresponding to the test T1-weighted image, and use a registration method to register the test segmentation image and the PDFF quantitative image to obtain the test fat quantification result. Use a registration method to register the liver mask test image of the test T1-weighted image and the PDFF quantitative image to obtain the standard fat quantification result. Calculate the difference between the test fat quantification result and the standard fat quantification result. The smaller the difference, the better the quality of the intermediate model.

[0159] For each intermediate model, input the multiple test T1-weighted images into each intermediate model in sequence for processing to obtain the test segmentation images of the multiple test T1-weighted images. Use a registration method to register each test segmentation image and the PDFF quantitative image to obtain multiple test fat quantification results. Use a registration method to register the liver mask test images of each test T1-weighted image and the PDFF quantitative image to obtain multiple standard fat quantification results.

[0160] Calculate the differences between the fat quantification results of each test and the fat quantification results of each standard to obtain a plurality of differences. Calculate the average value of the plurality of differences, and use this average value as the test result of each intermediate model.

[0161] S4043: Determine an image segmentation model among multiple intermediate models according to each test result.

[0162] Obtain the test results of each intermediate model. For example, obtain the average value of the differences of each intermediate model, and use the intermediate model with the smallest average value of the differences as the trained image segmentation model.

[0163] In this embodiment, by obtaining a sample test set to test multiple intermediate models, and selecting the intermediate model with the best quality among the multiple intermediate models as the trained image segmentation model according to the test results, the segmentation quality of the image segmentation model is guaranteed. Furthermore, when using this image segmentation model to perform rejection processing on the image corresponding to the liver blood vessels in the T1-weighted image later, an accurate liver segmentation image can be obtained.

[0164] Please refer to Figure 6 , Figure 6 is a schematic diagram of a device for segmenting images provided by an embodiment of the present application. Each unit included in the device for segmenting images is used to execute Figure 1 , Figure 2 , Figure 4 the respective steps in the corresponding embodiments. Specifically, please refer to Figure 1 , Figure 2 , Figure 4 the relevant descriptions in their respective corresponding embodiments. For the sake of convenience of description, only the parts related to this embodiment are shown. Refer to Figure 6 , including:

[0165] An acquisition unit 510, configured to acquire a T1-weighted image of the abdomen of a subject to be measured;

[0166] A processing unit 520, configured to perform rejection processing on the image corresponding to the liver blood vessels in the T1-weighted image through a trained image segmentation model to obtain a liver segmentation map.

[0167] Optionally, the T1-weighted image includes a T1-weighted in-phase image and a T1-weighted opposed-phase image, and the processing unit 520 is specifically configured to:

[0168] Extract the liver edge features in the T1-weighted opposed-phase image and the liver blood vessel features in the T1-weighted in-phase image through the image segmentation model;

[0169] Determine a liver image according to the liver edge features;

[0170] Determine the image corresponding to the liver blood vessels from the liver image according to the liver blood vessel characteristics;

[0171] Remove the image corresponding to the liver blood vessels from the T1-weighted image to obtain the liver segmentation image.

[0172] Optionally, the device further includes a registration unit for:

[0173] Obtain a quantitative image of the proton density fat fraction of the abdomen of the subject to be measured;

[0174] Register the liver segmentation image and the quantitative image of the proton density fat fraction to obtain a fat quantification result.

[0175] Optionally, the device further includes a training unit for:

[0176] Train the initial model using a preset sample training set and a loss function to obtain the trained image segmentation model. The sample training set includes a plurality of sample T1-weighted images, and a liver mask image corresponding to each sample T1-weighted image. The liver mask image is an image marked with the liver region and with the liver blood vessels removed.

[0177] Optionally, the training unit is further used for:

[0178] Input the sample T1-weighted images in the sample training set into the initial model for processing to obtain the sample segmentation images of the sample T1-weighted images;

[0179] Calculate the loss value between the sample segmentation image of the sample T1-weighted image and the liver mask image corresponding to the sample T1-weighted image according to the loss function;

[0180] When it is detected that the loss value is greater than a preset threshold, adjust the model parameters of the initial model, and continue to train the initial model using the sample training set;

[0181] When it is detected that the loss value is less than or equal to the preset threshold, stop training the initial model, and determine the trained initial model as the image segmentation model.

[0182] Optionally, the training unit is further used for:

[0183] Input the sample T1-weighted images in the sample training set into the initial model for processing to obtain the sample segmentation images of the sample T1-weighted images;

[0184] Calculate the loss value between the sample segmentation image of the sample T1-weighted image and the liver mask image corresponding to the sample T1-weighted image according to the loss function;

[0185] Output an intermediate model whenever it is detected that the loss value is less than or equal to a preset threshold until the number of training times reaches the preset number of training times, and obtain a plurality of the intermediate models;

[0186] Determine the image segmentation model from the plurality of intermediate models.

[0187] Optionally, the training unit is further configured to:

[0188] Obtain a sample test set, where the sample test set includes a plurality of test T1-weighted images and liver mask test images corresponding to each of the test T1-weighted images;

[0189] Test the plurality of intermediate models by using the sample test set;

[0190] Determine the image segmentation model from the plurality of intermediate models according to each test result.

[0191] Please refer to Figure 7 , Figure 7 which is a schematic diagram of a device for segmenting images provided in another embodiment of the present application. As Figure 7 shown, the device 6 for segmenting images in this embodiment includes: a processor 60, a memory 61, and a computer program 62 stored in the memory 61 and executable on the processor 60. When the processor 60 executes the computer program 62, the steps in the above-mentioned method embodiments for segmenting images are implemented, such as Figure 1 S101 to S102 shown. Alternatively, when the processor 60 executes the computer program 62, the functions of each unit in the above-mentioned embodiments are implemented, such as Figure 6 the functions of the units 510 to 520 shown.

[0192] Exemplarily, the computer program 62 can be divided into one or more units, and the one or more units are stored in the memory 61 and executed by the processor 60 to complete the present application. The one or more units can be a series of computer instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program 62 in the device 6 for segmenting images. For example, the computer program 62 can be divided into an acquisition unit and a processing unit, and the specific functions of each unit are as described above.

[0193] The device may include, but is not limited to, a processor 60 and a memory 61. Those skilled in the art can understand that Figure 7This is only an example of the device 6 for splitting images, which does not constitute a limitation on the device. It may include more or fewer components than those shown, or combine certain components, or different components. For example, the device may also include input / output devices, network access devices, buses, etc.

[0194] The so-called processor 60 may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0195] The memory 61 may be an internal storage unit of the device, such as the hard disk or memory of the device. The memory 61 may also be an external storage terminal of the device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the device. Further, the memory 61 may also include both the internal storage unit and the external storage terminal of the device. The memory 61 is used to store the computer instructions and other programs and data required by the terminal. The memory 61 may also be used to temporarily store the data that has been output or will be output.

[0196] The embodiments of the present application also provide a computer storage medium. The computer storage medium may be non-volatile or volatile. The computer storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the method embodiments for splitting images described above are implemented.

[0197] The present application also provides a computer program product. When the computer program product runs on a device, the device is caused to execute the steps in the method embodiments for splitting images described above.

[0198] The embodiments of the present application also provide a chip or integrated circuit. The chip or integrated circuit includes: a processor for calling and running a computer program from a memory, so that a device installed with the chip or integrated circuit executes the steps in the method embodiments for splitting images described above.

[0199] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above division of each functional unit and module is used as an example. In actual applications, the above functions can be assigned to different functional units and modules according to needs, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiments can be integrated into a processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit. In addition, the specific names of each functional unit and module are only for the convenience of mutual distinction and do not limit the protection scope of this application. The specific working processes of the units and modules in the above system can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated here.

[0200] In the above embodiments, the descriptions of the respective embodiments have their own emphases. For the parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0201] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or by a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0202] The above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit them; although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit scope of the technical solutions of each embodiment of this application and should all be included in the protection scope of this application.

Claims

1. A method for segmenting an image, characterized in that, Including: Obtain a T1-weighted image of the abdomen of the object to be measured; Through a trained image segmentation model, perform a removal process on the image corresponding to the liver blood vessels in the T1-weighted image to obtain a liver segmentation image; the training method of the trained image segmentation model is: use a preset sample training set and a loss function to train an initial model to obtain the trained image segmentation model, the sample training set includes multiple sample T1-weighted images, and the liver mask images corresponding to each of the sample T1-weighted images, and the liver mask image is an image with the liver region marked and the liver blood vessels removed.

2. The method according to claim 1, characterized in that, The T1-weighted image includes a T1-weighted in-phase image and a T1-weighted opposed-phase image. The step of performing a removal process on the image corresponding to the liver blood vessels in the T1-weighted image through a trained image segmentation model to obtain a liver segmentation image includes: Extract the liver edge features in the T1-weighted opposed-phase image and the liver blood vessel features in the T1-weighted in-phase image through the image segmentation model; Determine a liver image according to the liver edge features; Determine the image corresponding to the liver blood vessels from the liver image according to the liver blood vessel features; Remove the image corresponding to the liver blood vessels from the T1-weighted image to obtain the liver segmentation image.

3. The method according to claim 1, characterized in that, After the step of performing a removal process on the image corresponding to the liver blood vessels in the T1-weighted image through a trained image segmentation model to obtain a liver segmentation image, the method further includes: Obtain a proton density fat fraction quantitative image of the abdomen of the object to be measured; Register the liver segmentation image and the proton density fat fraction quantitative image to obtain a fat quantification result.

4. The method according to claim 1, characterized in that, The step of using a preset sample training set and a loss function to train an initial model to obtain the trained image segmentation model includes: Input the sample T1-weighted images in the sample training set into the initial model for processing to obtain a sample segmentation image of the sample T1-weighted images; Calculate the loss value between the sample segmentation image of the sample T1-weighted image and the liver mask image corresponding to the sample T1-weighted image according to the loss function; When it is detected that the loss value is greater than a preset threshold, adjust the model parameters of the initial model and continue to train the initial model using the sample training set; When it is detected that the loss value is less than or equal to the preset threshold, stop training the initial model and determine the trained initial model as the image segmentation model.

5. The method according to claim 1, characterized in that, The step of using a preset sample training set and a loss function to train an initial model to obtain the trained image segmentation model includes: Input the sample T1-weighted images in the sample training set into the initial model for processing to obtain a sample segmentation image of the sample T1-weighted images; Calculate the loss value between the sample segmentation image of the sample T1-weighted image and the liver mask image corresponding to the sample T1-weighted image according to the loss function; Output the intermediate model whenever it is detected that the loss value is less than or equal to a preset threshold until the number of training times reaches the preset number of training times, and obtain a plurality of the intermediate models; Determine the image segmentation model from the plurality of intermediate models.

6. The method according to claim 5, characterized in that, The determining the image segmentation model from the plurality of intermediate models includes: Obtain a sample test set, where the sample test set includes a plurality of test T1-weighted images and liver mask test images corresponding to each of the test T1-weighted images; Test the plurality of intermediate models using the sample test set; Determine the image segmentation model from the plurality of intermediate models according to each test result.

7. An apparatus for segmenting an image, characterized in that, Includes: An acquisition unit for acquiring a T1-weighted image of the abdomen of the object to be measured; A processing unit for removing the image corresponding to the liver blood vessels in the T1-weighted image through a trained image segmentation model to obtain a liver segmentation image; the training method of the trained image segmentation model is: training an initial model using a preset sample training set and a loss function to obtain the trained image segmentation model, the sample training set includes a plurality of sample T1-weighted images and liver mask images corresponding to each of the sample T1-weighted images, and the liver mask image is an image with the liver area marked and the liver blood vessels removed.

8. An image segmentation device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images